arXiv - QuanBio - Biomolecules最新文献

{"title":"Statistical Survey of Chemical and Geometric Patterns on Protein Surfaces as a Blueprint for Protein-mimicking Nanoparticles","authors":"John M. McBride, Aleksei Koshevarnikov, Marta Siek, Bartosz A. Grzybowski, Tsvi Tlusty","doi":"arxiv-2407.14063","DOIUrl":"https://doi.org/arxiv-2407.14063","url":null,"abstract":"Despite recent breakthroughs in understanding how protein sequence relates to\u0000structure and function, considerably less attention has been paid to the\u0000general features of protein surfaces beyond those regions involved in binding\u0000and catalysis. This paper provides a systematic survey of the universe of\u0000protein surfaces and quantifies the sizes, shapes, and curvatures of the\u0000positively/negatively charged and hydrophobic/hydrophilic surface patches as\u0000well as correlations between such patches. It then compares these statistics\u0000with the metrics characterizing nanoparticles functionalized with ligands\u0000terminated with positively and negatively charged ligands. These particles are\u0000of particular interest because they are also surface-patchy and have been shown\u0000to exhibit both antibiotic and anticancer activities - via selective\u0000interactions against various cellular structures - prompting loose analogies to\u0000proteins. Our analyses support such analogies in several respects (e.g.,\u0000patterns of charged protrusions and hydrophobic niches similar to those\u0000observed in proteins), although there are also significant differences. Looking\u0000forward, this work provides a blueprint for the rational design of synthetic\u0000nanoobjects with further enhanced mimicry of proteins' surface properties.","PeriodicalId":501022,"journal":{"name":"arXiv - QuanBio - Biomolecules","volume":"1 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-07-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141745125","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A foundation model approach to guide antimicrobial peptide design in the era of artificial intelligence driven scientific discovery","authors":"Jike Wang, Jianwen Feng, Yu Kang, Peichen Pan, Jingxuan Ge, Yan Wang, Mingyang Wang, Zhenxing Wu, Xingcai Zhang, Jiameng Yu, Xujun Zhang, Tianyue Wang, Lirong Wen, Guangning Yan, Yafeng Deng, Hui Shi, Chang-Yu Hsieh, Zhihui Jiang, Tingjun Hou","doi":"arxiv-2407.12296","DOIUrl":"https://doi.org/arxiv-2407.12296","url":null,"abstract":"We propose AMP-Designer, an LLM-based foundation model approach for the rapid\u0000design of novel antimicrobial peptides (AMPs) with multiple desired properties.\u0000Within 11 days, AMP-Designer enables de novo design of 18 novel candidates with\u0000broad-spectrum potency against Gram-negative bacteria. Subsequent in vitro\u0000validation experiments demonstrate that almost all in silico recommended\u0000candidates exhibit notable antibacterial activity, yielding a 94.4% positive\u0000rate. Two of these candidates exhibit exceptional activity, minimal\u0000hemotoxicity, substantial stability in human plasma, and a low propensity of\u0000inducing antibiotic resistance as observed in murine lung infection\u0000experiments, showcasing their significant efficacy in reducing bacterial load\u0000by approximately one hundredfold. The entire process, from in silico design to\u0000in vitro and in vivo validation, is completed within a timeframe of 48 days.\u0000Moreover, AMP-Designer demonstrates its remarkable capability in designing\u0000specific AMPs to target strains with extremely limited labeled datasets. The\u0000most outstanding candidate against Propionibacterium acnes suggested by\u0000AMP-Designer exhibits an in vitro minimum inhibitory concentration value of 2.0\u0000$mu$g/ml. Through the integration of advanced machine learning methodologies\u0000such as contrastive prompt tuning, knowledge distillation, and reinforcement\u0000learning within the AMP-Designer framework, the process of designing AMPs\u0000demonstrates exceptional efficiency. This efficiency remains conspicuous even\u0000in the face of challenges posed by constraints arising from a scarcity of\u0000labeled data. These findings highlight the tremendous potential of AMP-Designer\u0000as a promising approach in combating the global health threat of antibiotic\u0000resistance.","PeriodicalId":501022,"journal":{"name":"arXiv - QuanBio - Biomolecules","volume":"50 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-07-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141745049","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Learning Structurally Stabilized Representations for Multi-modal Lossless DNA Storage","authors":"Ben Cao, Tiantian He, Xue Li, Bin Wang, Xiaohu Wu, Qiang Zhang, Yew-Soon Ong","doi":"arxiv-2408.00779","DOIUrl":"https://doi.org/arxiv-2408.00779","url":null,"abstract":"In this paper, we present Reed-Solomon coded single-stranded representation\u0000learning (RSRL), a novel end-to-end model for learning representations for\u0000multi-modal lossless DNA storage. In contrast to existing learning-based\u0000methods, the proposed RSRL is inspired by both error-correction codec and\u0000structural biology. Specifically, RSRL first learns the representations for the\u0000subsequent storage from the binary data transformed by the Reed-Solomon codec.\u0000Then, the representations are masked by an RS-code-informed mask to focus on\u0000correcting the burst errors occurring in the learning process. With the decoded\u0000representations with error corrections, a novel biologically stabilized loss is\u0000formulated to regularize the data representations to possess stable\u0000single-stranded structures. By incorporating these novel strategies, the\u0000proposed RSRL can learn highly durable, dense, and lossless representations for\u0000the subsequent storage tasks into DNA sequences. The proposed RSRL has been\u0000compared with a number of strong baselines in real-world tasks of multi-modal\u0000data storage. The experimental results obtained demonstrate that RSRL can store\u0000diverse types of data with much higher information density and durability but\u0000much lower error rates.","PeriodicalId":501022,"journal":{"name":"arXiv - QuanBio - Biomolecules","volume":"104 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-07-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141942275","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Directly Optimizing for Synthesizability in Generative Molecular Design using Retrosynthesis Models","authors":"Jeff Guo, Philippe Schwaller","doi":"arxiv-2407.12186","DOIUrl":"https://doi.org/arxiv-2407.12186","url":null,"abstract":"Synthesizability in generative molecular design remains a pressing challenge.\u0000Existing methods to assess synthesizability span heuristics-based methods,\u0000retrosynthesis models, and synthesizability-constrained molecular generation.\u0000The latter has become increasingly prevalent and proceeds by defining a set of\u0000permitted actions a model can take when generating molecules, such that all\u0000generations are anchored in \"synthetically-feasible\" chemical transformations.\u0000To date, retrosynthesis models have been mostly used as a post-hoc filtering\u0000tool as their inference cost remains prohibitive to use directly in an\u0000optimization loop. In this work, we show that with a sufficiently\u0000sample-efficient generative model, it is straightforward to directly optimize\u0000for synthesizability using retrosynthesis models in goal-directed generation.\u0000Under a heavily-constrained computational budget, our model can generate\u0000molecules satisfying a multi-parameter drug discovery optimization task while\u0000being synthesizable, as deemed by the retrosynthesis model.","PeriodicalId":501022,"journal":{"name":"arXiv - QuanBio - Biomolecules","volume":"26 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-07-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141745124","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Repurformer: Transformers for Repurposing-Aware Molecule Generation","authors":"Changhun Lee, Gyumin Lee","doi":"arxiv-2407.11439","DOIUrl":"https://doi.org/arxiv-2407.11439","url":null,"abstract":"Generating as diverse molecules as possible with desired properties is\u0000crucial for drug discovery research, which invokes many approaches based on\u0000deep generative models today. Despite recent advancements in these models,\u0000particularly in variational autoencoders (VAEs), generative adversarial\u0000networks (GANs), Transformers, and diffusion models, a significant challenge\u0000known as textit{the sample bias problem} remains. This problem occurs when\u0000generated molecules targeting the same protein tend to be structurally similar,\u0000reducing the diversity of generation. To address this, we propose leveraging\u0000multi-hop relationships among proteins and compounds. Our model, Repurformer,\u0000integrates bi-directional pretraining with Fast Fourier Transform (FFT) and\u0000low-pass filtering (LPF) to capture complex interactions and generate diverse\u0000molecules. A series of experiments on BindingDB dataset confirm that\u0000Repurformer successfully creates substitutes for anchor compounds that resemble\u0000positive compounds, increasing diversity between the anchor and generated\u0000compounds.","PeriodicalId":501022,"journal":{"name":"arXiv - QuanBio - Biomolecules","volume":"10 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-07-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141721577","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Unzipping of knotted DNA via nanopore translocation","authors":"Antonio Suma, Cristian Micheletti","doi":"arxiv-2407.11567","DOIUrl":"https://doi.org/arxiv-2407.11567","url":null,"abstract":"DNA unzipping by nanopore translocation has implications in diverse contexts,\u0000from polymer physics to single-molecule manipulation to DNA-enzyme interactions\u0000in biological systems. Here we use molecular dynamics simulations and a\u0000coarse-grained model of DNA to address the nanopore unzipping of DNA filaments\u0000that are knotted. This previously unaddressed problem is motivated by the fact\u0000that DNA knots inevitably occur in isolated equilibrated filaments and in vivo.\u0000We study how different types of tight knots in the DNA segment just outside the\u0000pore impact unzipping at different driving forces. We establish three main\u0000results. First, knots do not significantly affect the unzipping process at low\u0000forces. However, knotted DNAs unzip more slowly and heterogeneously than\u0000unknotted ones at high forces. Finally, we observe that the microscopic origin\u0000of the hindrance typically involves two concurrent causes: the topological\u0000friction of the DNA chain sliding along its knotted contour and the additional\u0000friction originating from the entanglement with the newly unzipped DNA. The\u0000results reveal a previously unsuspected complexity of the interplay of DNA\u0000topology and unzipping, which should be relevant for interpreting\u0000nanopore-based single-molecule unzipping experiments and improving the modeling\u0000of DNA transactions in vivo.","PeriodicalId":501022,"journal":{"name":"arXiv - QuanBio - Biomolecules","volume":"19 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-07-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141721576","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A Hitchhiker's Guide to Deep Chemical Language Processing for Bioactivity Prediction","authors":"Rıza Özçelik, Francesca Grisoni","doi":"arxiv-2407.12152","DOIUrl":"https://doi.org/arxiv-2407.12152","url":null,"abstract":"Deep learning has significantly accelerated drug discovery, with 'chemical\u0000language' processing (CLP) emerging as a prominent approach. CLP learns from\u0000molecular string representations (e.g., Simplified Molecular Input Line Entry\u0000Systems [SMILES] and Self-Referencing Embedded Strings [SELFIES]) with methods\u0000akin to natural language processing. Despite their growing importance, training\u0000predictive CLP models is far from trivial, as it involves many 'bells and\u0000whistles'. Here, we analyze the key elements of CLP training, to provide\u0000guidelines for newcomers and experts alike. Our study spans three neural\u0000network architectures, two string representations, three embedding strategies,\u0000across ten bioactivity datasets, for both classification and regression\u0000purposes. This 'hitchhiker's guide' not only underscores the importance of\u0000certain methodological choices, but it also equips researchers with practical\u0000recommendations on ideal choices, e.g., in terms of neural network\u0000architectures, molecular representations, and hyperparameter optimization.","PeriodicalId":501022,"journal":{"name":"arXiv - QuanBio - Biomolecules","volume":"35 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-07-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141744877","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Antibody DomainBed: Out-of-Distribution Generalization in Therapeutic Protein Design","authors":"Nataša Tagasovska, Ji Won Park, Matthieu Kirchmeyer, Nathan C. Frey, Andrew Martin Watkins, Aya Abdelsalam Ismail, Arian Rokkum Jamasb, Edith Lee, Tyler Bryson, Stephen Ra, Kyunghyun Cho","doi":"arxiv-2407.21028","DOIUrl":"https://doi.org/arxiv-2407.21028","url":null,"abstract":"Machine learning (ML) has demonstrated significant promise in accelerating\u0000drug design. Active ML-guided optimization of therapeutic molecules typically\u0000relies on a surrogate model predicting the target property of interest. The\u0000model predictions are used to determine which designs to evaluate in the lab,\u0000and the model is updated on the new measurements to inform the next cycle of\u0000decisions. A key challenge is that the experimental feedback from each cycle\u0000inspires changes in the candidate proposal or experimental protocol for the\u0000next cycle, which lead to distribution shifts. To promote robustness to these\u0000shifts, we must account for them explicitly in the model training. We apply\u0000domain generalization (DG) methods to classify the stability of interactions\u0000between an antibody and antigen across five domains defined by design cycles.\u0000Our results suggest that foundational models and ensembling improve predictive\u0000performance on out-of-distribution domains. We publicly release our codebase\u0000extending the DG benchmark ``DomainBed,'' and the associated dataset of\u0000antibody sequences and structures emulating distribution shifts across design\u0000cycles.","PeriodicalId":501022,"journal":{"name":"arXiv - QuanBio - Biomolecules","volume":"16 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-07-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141870339","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Structural Dynamics of Contractile Injection Systems","authors":"Noah Toyonaga, L Mahadevan","doi":"arxiv-2407.10291","DOIUrl":"https://doi.org/arxiv-2407.10291","url":null,"abstract":"The dynamics of many macromolecular machines is characterized by\u0000chemically-mediated structural changes that achieve large scale functional\u0000deployment through local rearrangements of constitutive protein sub-units.\u0000Motivated by recent high resolution structural microscopy of a particular class\u0000of such machines, contractile injection systems (CIS), we construct a coarse\u0000grained semi-analytical model that recapitulates the geometry and bistable\u0000mechanics of CIS in terms of a minimal set of measurable physical parameters.\u0000We use this model to predict the size, shape and speed of a dynamical actuation\u0000front that underlies contraction. Scaling laws for the velocity and physical\u0000extension of the contraction front are consistent with our numerical\u0000simulations, and may be generally applicable to related systems.","PeriodicalId":501022,"journal":{"name":"arXiv - QuanBio - Biomolecules","volume":"30 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-07-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141722539","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Any-Property-Conditional Molecule Generation with Self-Criticism using Spanning Trees","authors":"Alexia Jolicoeur-Martineau, Aristide Baratin, Kisoo Kwon, Boris Knyazev, Yan Zhang","doi":"arxiv-2407.09357","DOIUrl":"https://doi.org/arxiv-2407.09357","url":null,"abstract":"Generating novel molecules is challenging, with most representations leading\u0000to generative models producing many invalid molecules. Spanning Tree-based\u0000Graph Generation (STGG) is a promising approach to ensure the generation of\u0000valid molecules, outperforming state-of-the-art SMILES and graph diffusion\u0000models for unconditional generation. In the real world, we want to be able to\u0000generate molecules conditional on one or multiple desired properties rather\u0000than unconditionally. Thus, in this work, we extend STGG to\u0000multi-property-conditional generation. Our approach, STGG+, incorporates a\u0000modern Transformer architecture, random masking of properties during training\u0000(enabling conditioning on any subset of properties and classifier-free\u0000guidance), an auxiliary property-prediction loss (allowing the model to\u0000self-criticize molecules and select the best ones), and other improvements. We\u0000show that STGG+ achieves state-of-the-art performance on in-distribution and\u0000out-of-distribution conditional generation, and reward maximization.","PeriodicalId":501022,"journal":{"name":"arXiv - QuanBio - Biomolecules","volume":"54 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-07-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141721579","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}